[pulseaudio-discuss] [PATCH v2] pacat: Write to stream in frame-sized chunks

[Arun Raghavan]

On Tue, 20 Dec 2016, at 03:14 PM, Ahmed S. Darwish wrote:
> Hi Tanu,
> 
> On Sun, Dec 18, 2016 at 08:59:58PM +0200, Tanu Kaskinen wrote:
> > On Sat, 2016-12-17 at 01:43 +0200, Ahmed S. Darwish wrote:
> > > # Note: Documentation of `pa_stream_begin_write(stream, &buf, &nbytes)'
> > > # suggests to pass nbytes = -1. Doing so though lends to a huge returned
> > > # nbytes values. Calling a pa_stream_write() with such values chokes the
> > > # stream: making it go silent and not produce any further write-callback
> > > # events.
> > > #
> > > # Is this a problem in usage pattern, or a bug?
> > 
> > Sounds like a bug. Trying to write more than buffer_attr.maxlength
> > bytes at once would be expected to cause some kind of problems, but I
> > don't think that's the case here. The returned nbytes value is 64k,
> > isn't it? pacat uses the default maxlength, and the default maxlength
> > is 4M, so nbytes is nowhere near that limit. As long as maxlength isn't
> > exceeded, any "extra" audio should just get buffered in the stream
> > buffer at the server side.
> >
> 
> Yeah; sounds like a bug in the write stream indeed.
> 
> In the command below, (second pacat):
> 
>   ./src/pacat -r --latency-msec=4 | ./src/pacat --latency-msec=4
> 
> When using "writable" values from pa_stream_writable_size(), we
> get the following, valid, sequence:
> 
>   ++ STDIN callback: mute! stream writable = 0
>   @@ WRITE callback: Re-enable STDIN events
>   <-- Small writable val; appropriate for requested small latency -->  
>   @@ STDIN callback: stream writable size = 192
>   ** Writing 192 bytes; frame len = 4
> 
>   ++ STDIN callback: mute! writable = 0
>   @@ WRITE callback: Re-enable STDIN events
>   <-- Small writable val; appropriate for requested small latency -->  
>   @@ STDIN callback: stream writable size = 192
>   ** Writing 192 bytes; frame len = 4
> 
>   ... Sequence endlessly repeated ...
> 
> BUT when using "writable" values from pa_stream_begin_read(), with
> "writable = (size_t)-1" as the documentation recommends, we get
> that sequence instead:
> 
>   @@ WRITE callback: Re-enable STDIN events
>   @@ STDIN callback: stream writable size = 65472   <== 64K indeed
>   ** Writing 2112 bytes; frame len = 4
>   ++ STDIN callback: mute! writable = 0
> 
>   <-- COMPLETE silence! - no more stream write callback events -->
> 
> Hopefully not a new blocker, but if you agree with the assessment
> above, I'll submit a bug to be tracked for v11.

Is that a regression?

-- Arun